Lane detection and traffic sign recognition

Abstract

In this paper, some image processing techniques are proposed to detect lanes, and Convolutional Neural Networks (CNN) to detect and classify traffic signs as well. As for the lane detection the main steps are getting the region of interest (ROI) from the input image, thresholding this image, segment lane pixels, and draw the created lane on to the main image. For the traffic sign detection a Yolov3 model was used, trained by darknet framework with the GTSDB database, and for the classification, three CNN models were evaluated, which were trained with the GTSRB database. With the GTSRB database, some image augmentation techniques were used to generate more images, so the database would be balanced.